"Heartbreaking: The Worst Person You Know Just Made a Great Point" – ClickHole "When the facts change, I change my mind. What do you do, sir?" – John Maynard Keynes, paraphrased 2025 was a weird year for me. If you had asked me exactly a year ago, I would have said I thought LLMs were…

Read the Tea Leaves

A developer's perspective shift on AI coding tools over 2025, moving from skepticism to having 90% of code authored by Claude Code. The piece examines how debates around LLMs have become tribal and political, acknowledging both the genuine capabilities of modern AI coding assistants (bug detection, automated fixes, iterative improvements) and the uncertainty about software development's future. The author advocates for experimentation and empathy rather than entrenched positions, noting that while the technology has crossed a practical threshold, nobody truly knows where it's heading.

AI tribalism

Imagine that AI coding was invented in the era of assemby language. Then all future development would be made in assembly with help of AI coding, as it would be good enough.
AI is a technology that tangles technology. We spend too much time on productive work than on creative research.
But we need research. Lets take example of JavaScript. We all agree that we need something less messy, less temporary, that would become a web standard. But with AI we dont bother, we dont do research, just stacking existing temporary solutions, creating even more mess.
Back then StackOverflow was a place of research, free experimentation, new questions, new ideas. Now its dead. They just put SO in the box and they sell this box to you as a best technology.
Research in software engineering is dead thanks to AI. Or rather thanks to how AI is meaninglessly shake all tech sectors. Im really pro AI if it is treated fairly as just an encapsulation of existing technology.

Love this shift from skepticism to measured collaboration with AI. It’s refreshing to hear a realistic take, Claude Code isn’t a magic wand, but a powerful teammate for bug hunts, quick refactors, and iteration loops. The tribal debates around lmss have missed how much we actually gain from nudging and validating AI suggestions together .humans guiding the brain, ai handling the heavy lift of repetitive fixes and scanning edge cases! been using claude for a lot of itme now